How science is published

Filters, incentives and their failures.

Peer review

Specialists check a manuscript for obvious flaws before publication. They rarely see raw data and cannot detect fraud, so review is a filter rather than a guarantee.

Preprints share work before review, faster and less filtered.
Retractions exist and are rarer than the error rate implies.

Publication bias

Positive, novel results are published far more readily than null ones, so the literature looks more certain than the evidence. Registered reports, accepted before results are known, are one fix.